Intensity Inhomogeneity Suppressed Fuzzy C-Means for Infrared Pedestrian Segmentation

Pedestrians are highlighted regions in infrared images. Hence, infrared images could be used for pedestrian detection. Segmentation is an important step for detection and an accurate segmentation would be helpful for infrared pedestrian detection. However, intensity inhomogeneity is a common drawback in infrared images due to occlusion or uneven heat dissipation. This phenomenon would have negative influences on pedestrian segmentation in infrared images. To address this problem, an intensity inhomogeneity suppressed fuzzy C-means method is proposed in this paper for the segmentation of infrared pedestrians with intensity inhomogeneity. Two improvements are made in the proposed method to suppress intensity inhomogeneity: 1) a weight based on membership information and image intensity information of infrared pedestrians is added to the objective function and 2) neighborhood information is considered by adding a regularization term into the objective function to suppress the intensity inhomogeneity in infrared images. Eight famous segmentation methods are utilized as comparison methods in our experiment. The experimental results show that the proposed method could effectively suppress the intensity inhomogeneity in infrared pedestrian images and perform better for segmentation than the comparison methods.